AT394295B - INTERFACE DEVICE FOR CONNECTING TELEPHONE LINES TO A DIGITAL SWITCH - Google Patents

Description

AT 394 295 B

The invention relates to an interface device for connecting analog, full-duplex remote telephone lines to a digital exchange, which consists of a group of two-wire four-wire converters, each of which is connected to a single remote line and an analog-to-digital converter in the four-wire reception path and contains a digital-to-analog converter in the four-wire transmission path and digital filters, the converters effecting a resistance adaptation to the subscriber lines.

In known devices for two-wire-four-wire implementation, the problem of false mismatch in the central telephone exchange between the subscriber lines or connecting lines and the terminating impedance is well known. Such mismatch causes poor echo cancellation properties and reflections or echoes due to the inadequate operation of the two-wire, four-wire hybrid circuit, which, in order to function ideally, requires that the line impedance and terminating impedance be magnitude and phase-wise across the bandwidth of the remote channel are the same. Passive fixed termination impedances are a compromise, apart from a specific frequency, because such a compromise impedance is either a series connection or a parallel connection of a resistor and a capacitor. Such a simple impedance results in an insufficient adaptation to the line impedance. The hybrid circuit that performs the two-wire to four-wire conversion is dependent on a relatively close match between the line impedance and the terminating impedance, and good impedance matching over the frequency range of interest is required if the hybrid circuit is to have optimal properties. Compensation for the mismatch has been attempted in known hybrid circuits with replica or supplementary networks which are part of the hybrid circuit and which represent a special configuration of the hybrid circuit for a given line. Another problem that arises from impedance mismatch at the far end of the line arises when this end is improperly terminated. An unwanted reflection or echo then comes back to the near end. If the terminating impedance at the near end is equal to the line impedance and if the fork transition loss is zero, the circuit at the near end has optimal properties. The far end echo can also be minimized using known echo cancellation techniques.

This known echo compensation depends on the fact that the locally transmitted signal is known from the start and that an uncorrelated relationship between the signal transmitted at the near end and the signal received from the far end is assumed.

Using the technique of adaptive equalization, with or without training signals, the correlated portion of the signal reflected at the far end contained in the signal received at the near end can be regenerated and subtracted from this locally received signal.

The adaptive equalizer has to meet certain conditions in order to work properly. The equalizer must have an adequate signal-to-noise ratio to allow convergence and it must have a linear network characteristic. Sufficient energy must be available over the frequency band to allow correct feedback control signals to adjust the coefficients of the digital filter. No crosstalk is permitted during the equalization process. The next and most significant difficulty is that the local digital telephone exchange that has to cooperate with analog subscriber lines must now add two-wire to four-wire converters to be able to cooperate with these lines. In the past, analog switchboards did not require hybrid switching.

These recently introduced interfaces in the form of hybrid circuits can cause reflections or unwanted reflux signals. Since these additional hybrid circuits were not previously available at the analog exchanges, the digital exchange may be weaker in performance than its predecessor made in analog technology if the properties of the hybrid circuit themselves are not improved.

The problem of whistling or, more precisely, the possible instability of the network in the Nyquist sense is the result of the undesired feedback that arises in the two-wire-four-wire implementation. As a result, the system tends to vibrate if no suitable measures are taken.

The attenuation plan takes this state into account in a conventional manner, in which suitable attenuations are inserted in the prescribed manner distributed over the telephone network and in which it is stipulated that at the points in the network where a two-wire-four-wire conversion takes place, the fork transition loss meets certain minimum criteria should.

Previously, in the context of the attenuation that could be inserted into existing networks, those switches (or circuits) that used two-wire to four-wire converters were considered, namely the connecting lines. For local exchanges in analog technology that do not require two-wire to four-wire converters, zero attenuation was allowed, and the permissible insertion loss was and is only a few tenths of a dB. Thus, the problem of designing a local exchange in digital technology with equivalent performance characteristics in an environment of analog technology is complicated by the existing plan. Experimental results show that the addition of an attenuation (4 dB) in the local exchange to solve this problem only leads to a reduction in the quality of operation, i. that is, the subscribers notice the reduced performance due to the added attenuation -2-

AT 394 295 B. Automatic equalizers per se are well known in the field of digital data transmission, for example from US Pat. No. 3,579,109 and from US Pat. No. 3,984,789. An adaptive digital equalizer is described in US Pat. No. 3,633,105 and the US -PS 3 789 560 describes an adaptive transversal equalizer that uses a second degree digital filter in time division multiplexing.

From DE-OS 2 808 737 an interface device between a bidirectional subscriber channel with analog signals and a digital device, in particular subscriber circuit for fem-intercoms with a digital switching matrix is known, in which an encoder converting analog signals into digital signals is provided, the output of which is provided with an attenuation -Device is connected, which performs a selective damping of the output signal of the encoder by subtracting unwanted signals in digital form, the output of the damping device being connected to the digital device, a decoder converting the digital signals arriving from the digital device into analog signals is provided is, the output of which is connected to a summing circuit via which the analog signal is transmitted to the subscriber channel, and the digital signals supplied to the decoder are supplied in parallel to the damping device and as undesired signals from Output signal of the encoder can be subtracted.

This subscriber line circuit still contains essential circuits in analog technology, such as. B. the analog filter and the summing circuit, so that only part of the entire circuit can be economically integrated.

It is therefore an object of the invention to provide a circuit arrangement in which largely digital signal processing takes place in the two-wire to four-wire converter, which effects an impedance adaptation to the impedance of the telephone line, which decouples and equalizes the signals received by the fem-speaking line and which can be manufactured using highly integrated circuit technology. This object is achieved according to the invention in that in the two-wire-four-wire converter the one input of a digital summing circuit is connected to the output of the analog-digital converter in the four-wire reception path, which converts the analog signal received by the telephone line into a composite digital signal is and the other input is connected to the transmit input of the digital exchange, the digital output signal of the analog-to-digital converter being added as a negative feedback signal to the digital transmit signal coming from the digital exchange that the output of the digital summing circuit with the input of a recursive digital filter with adjustable filter coefficients for the direct setting of a predetermined output impedance curve is connected that the output of this digital filter is connected to the input of the digital-to-analog converter, whose analog output signal voltage, which the over the long distance ng to be emitted information signal is supplied to a device for converting the analog signal voltage into an analog current, which is applied to the telephone line and to the input of the analog-digital converter, this device having a high output impedance compared to the adapted telephone line and with the impedance set by the digital filter, that the interface device further includes a central part common to the group of two-wire-four-wire converters, and that the central part contains a multiplexer and a filter coefficient adjusting unit, the multiplexer connecting the filter coefficient adjusting unit to each of the two-wire Turns on the four-wire converter of the group in time-division multiplexing mode and the filter coefficient adjustment unit by minimizing an error between the digital received signal which the two-wire four-wire converter transmits to the digital exchange and ei simulates the filter coefficients for automatic impedance matching.

The circuit according to the invention has the advantage that it assembles the output impedance digitally and thus works as an equalization circuit, whereby it manages with a simple recursive filter circuit, as described, for example, in US Pat. No. 3,716,807. The digital subscriber line circuit dispenses with discrete analog components. The functions of the complete electronic hybrid circuit and the digital impedance matching are implemented in a highly integrated circuit, and without extensive analogue adjustment or setting circuits.

According to one embodiment of the invention, the device for converting the analog signal voltage into an analog current consists of a steepness amplifier. Such an amplifier has an infinitely high output impedance and can be easily manufactured using FET technology and thus integrated together with other modules. It is thus possible to implement a complete subscriber line circuit on a single, highly integrated module.

A further development of the invention provides that the filter coefficients of the digital filter are set in an adaptive manner by a filter coefficient adjustment unit. Once the filter coefficients are set, they only need to be checked periodically since any given telephone line characteristic does not normally change from one day to the next. This allows the coefficient adjuster to be used by a plurality of subscriber line circuits in time-division multiplexing, so that their cost is divided over a number of lines.

According to a development of the invention, the recursive digital filter has five adjustable filter coefficients and thus leads to a considerable simplification of the circuit complexity compared to non-recursive structures.

A further embodiment of the invention is characterized in that in each two-wire four-wire -3-

AT 394 295 B

A second digital filter with adjustable coefficients is connected between the coding path and the decoding path, to which the digital transmission signal coming from the digital exchange is fed and which, by multiplying its input signal by the coefficients, outputs a signal representing the undesired portion of the reception signal originating from the transmission signal, whereby the adjustable coefficients of this filter are derived from the filter coefficients set in the filter coefficient adjustment unit, and that a digital summing circuit combines the output signal of the second digital filter with the composite digital output signal of the analog-digital converter in such a way that its undesirable echo components originating from the transmission signal be compensated.

According to a further embodiment of the invention, the second digital filter is a recursive digital filter. Such filters require a significantly smaller number of coefficients.

According to a further development of the invention, the analog-digital converter is a sigma-delta encoder and has a pre-filter and a post-filter. The pre-filter serves to limit its input bandwidth and the post-filter serves to set a desired feedback amplifier curve. This prevents spectral folding and also keeps the signal-to-noise ratio of the analog-digital converter as large as possible.

In a further embodiment of the invention, a switch controlled by the central part is provided in the feedback path between the output of the analog-digital converter and the digital summing circuit upstream of the first digital filter. The desired line termination impedance is formed in a digital manner by the digital loop from the coding path to the decoding

In practice, the central part is advantageously implemented in the form of a single, highly integrated circuit and each two-wire to four-wire converter in the form of a single, highly integrated circuit.

A further embodiment of the invention is characterized in that the filter coefficient adjustment unit calculates a digital gradient vector with the components CQ to C ^ by digitally processing the input signal and the reference signal, and the coefficients of the recursive filter depending on the calculated gradient vector in a rapidly converging process adapüv adjusts so that the digital filter works recursively and emits an equalized output signal after convergence.

The invention will now be explained, for example, with reference to the drawings. Show it:

1 is a known electronic hybrid circuit,

2 shows an interface device according to the invention,

3a shows the principle of the digital setting of an output admittance,

3b shows a simplified block diagram for adjusting impedance in a digital hybrid circuit, FIG. 4 shows an exemplary embodiment of a steepness amplifier,

5 shows the principle of the filter coefficient adjustment unit of an automatic recursive equalizer,

6 is a block diagram of a generalized digital filter,

7 is a block diagram of a feed coefficient adjustment unit for an equalizer.

8 is a block diagram of a generalized two-wire to four-wire converter in digital technology,

9 is a block diagram of a preferred embodiment of a filter coefficient adjustment unit according to the invention,

10 shows a first memory part of the adjusting unit according to FIG. 9,

11 shows a second memory part of the adjusting unit according to FIG. 9,

12 is a flowchart of the filter coefficient adjustment process.

13 shows the multiplexer used according to the invention and

14 shows the interface connections between the filter coefficient adjustment unit according to FIG. 9 and the multiplexer used according to the invention.

1 shows, designated (10), a known electronic hybrid circuit which performs the following functions: (1) adaptation to the input impedance of the line at its connections (12 and 14), which in FIG. 1 is a concentrated impedance Z ^ (16) is shown; (2) Minimization of the unwanted echo signal from the transmission path (1) by providing a compensation signal on the transmission path (2) in such a way that the transmission signal is reduced or minimized at the output of the reception amplifier (18). The hybrid circuit (10) works as follows:

The impedances Zg (20), Z ^ and Z ^ (22) and Z ^ (24) form a bridge circuit in which a

Transmitting amplifier (26) represents the voltage source and a receiving amplifier (18) acts as the zero instrument. When the bridge is balanced, the output of the receive amplifier (18) is zero for each transmit signal. At the same time, each signal appearing on the wires a and b is received at the output of the amplifier (18). Undesired backflows from the transmission signal are thus eliminated, so that the hybrid circuit effects the two-wire-four-wire conversion. The impedance seen by Z ^ is Zg and the parallel -4-

AT 394 295 B

Impedance Zj (28). If Zj is much greater than Zg, then for the transmission line the line termination resistance is Zg. In practice Zg is a fixed value, typically 900 ohms in series with 2.2 microfarads, Zj and Z4 are changed or selected so that the suitable adjustment takes place and thereby the unwanted reflux signal is minimized.

The disadvantages of this circuit are as follows: (1) Since Zg is not equal to Z ^ on all transmission lines, the return loss of circuit (10) is defined by:

Zl + Zg return loss = 20 logjQ- (Eq. 1) ZL'Z8 not infinite, which is the optimal value for minimizing reflections from signals received from the far end. (2) Z3 and Z4 must be selected on a compromise basis that the circuit is suitable for different lines, or Z3 and Z4 must be set manually or automatically by an appropriate set of control signals. These control signals then activate an analog switching network which switches suitable values of Z3 and Z4 into the circuit in accordance with a control process. (3) The majority of the impedances shown in the circuit of FIG. 1 are complex impedances and therefore require both resistors and capacitors. (4) The circuit according to FIG. 1 is basically of an analog type and is not suitable for production in inexpensive, economical and highly integrated digital technology. (5) The circuit of FIG. 1 requires expensive, stable, and highly accurate components to function reliably during the life of the device with which it is used. (6) Input and output of the four-wire side of the circuit of FIG. 1 are the decoders and coders in the codec used in a digital switching system. Thus, in order to function well, the subscriber line circuit for an exchange in digital technology must contain its own codec and fork circuit, in addition to the normal codec filters.

Fig. 2 shows, generally designated (100), a block diagram of a digital subscriber line circuit according to the invention, the function of a two-wire-four-wire hybrid circuit being carried out without the previously used discrete analog components. The circuit (100) also forms a line terminating impedance for each transmission line characteristic, which is automatically adapted in a digital manner. These features enable the construction of a highly integrated (LSI) circuit, which includes the complete functions of the electronic hybrid circuit and the impedance matching as part of an overall codec function, these functions being implemented on a single highly integrated module (LSI) without extensive external adjustment - or setting circuits in analog technology. This results in a low-cost, long-distance, high-density, high-reliability remote telephone subscriber line circuit. The technique on which this circuit is based is described below.

In Fig. 1, if the impedance Zg could be made exactly equal to the impedance Z ^ over the range of transmission lines for which the impedance is intended, then the impedances Z3 and Z4 could be made ohmic resistances and there would be no return signals and the impedance matching becomes so precise that the return loss, as described above, would be maximum. The circuit of FIG. 2 achieves this by using digital filtering and feedback methods without using the previously used analog components. 2, the output impedance of the circuit (100), viewed between the a and b wires (102) and (104), is designed such that it is matched to the input impedance of the line.

The line terminating impedance Zg, which is mentioned in connection with FIG. 1, is formed by the digital loop of the circuit (100) in a digital manner when a switch (106) is closed, which via a line (112) a feedback path from Establishes encoder path (108) to the decoding path (110).

3B shows a simplified block diagram of such a digital formation of the impedance. A steepness amplifier or g-amplifier (200) is a circuit arrangement which converts an input voltage Vx into an output current -gVx. It has an infinite output impedance or zero output conductance. This type of circuit arrangement is well known per se. In its -5-

AT 394 295 B simplest form, it can be a pentode vacuum tube or a field-effect transistor, the anode current or the drain current being proportional to the grid voltage or the gate-drain voltage. In the exemplary embodiment denoted by (200) in FIG. 4, an operational amplifier with high open-loop gain (gain equal to 10 **) is used, with a resistor whose conductance is equal to Siemens. In practice, the amplifier (200) is the normal output amplifier of the Decoder (114) and does not represent an additional component. It is shown as functionally separate only for the purpose of explanation. The decoder (114) is a circuit arrangement that converts a digital signal into an analog voltage, and this function can be realized with known digital-to-analog converters. The H filter is a digital filter with programmable coefficients, as is well known in the context of digital filtering and will be explained below. A digital summing circuit (118) (FIG. 3B) is connected upstream of the H filter (116). The pre-filter (120) serves to adjust the bandwidth of the input signal of the decoder (114) to the area of interest, i. H. to the frequency range below the sampling frequency of the encoder (122), in order to avoid spectral convolution and to make the signal-to-noise ratio of the encoder (122) as large as possible. The pre-filter (120) can consist of a simple passive two-pole low-pass filter, since it has to process analog signals.

The encoder (122) is preferably a sigma-delta encoder as described by Candy et al. The post-filter (124) causes a bandwidth limitation and, in combination with the pre-filter (120), ensures an overall flat low-pass filter curve, so that the composite gain in the frequency band of interest is equal to 1 and drops off monotonously outside this band.

In the arrangement defined in this way, the following applies to the frequency response at the points indicated in FIG. 3B:

Vo ’-gH - = -, where YL '= YL + Ybf + Y f (Eq. 2)

Vin YL ’+ gH

The output admittance can be represented as:

Y off

= YL ’+ gH (Eq. 3), taking into account the battery supply (125), the prefilter (120) and the line loads.

In Fig. 3A, the digital formation of an output admittance is shown with a simplified and generalized circuit that can be used for a variety of applications in which a specific output admittance is to be generated digitally and in which other analog-digital and digital -Analog converter can be used as the one specifically intended for telephone equipment. The circuit can also be used as an encoder / decoder. The pre-filter (250) has the function of limiting the bandwidth of the input signal Vin on the line (252) in order to prevent spectral convolution due to sampling with a sampling frequency which is too low. After the analog-to-digital conversion in the analog-to-digital converter, the post-filter (254) provides overall a low-pass characteristic of the analog-to-digital converter path (256) with a gain factor 1. The g-amplifier (200) and that digital H filters (116) are described elsewhere. The pre-filter (258) of the digital-to-analog converter provides an amplification factor equal to gH in the digital-to-analog converter path (260).

The post-filter (262) eliminates or minimizes the quantization effects occurring in the digital-to-analog converter (264). Quantization effects are defined as the noise that occurs as a result of an error in approaching a continuous analog output signal. In the event that the impedances of the pre-filter and the battery supply are large compared to the impedance of the line, they can be neglected (in other words, if the conductance values are small compared to the conductance value of the line, they can be neglected). Typically, the battery supply is designed to have this property to prevent signal loss and coupling of unwanted signals in the codec. The pre-filter can also be designed to have this property when using a large series resistance or the large input impedance of the gate input of a field effect transistor

Under these conditions, Y ^ '= Y and the output admittance is:

Y off

= YL + gH (Eq. 4). -6-

AT 394 295 B

If gH is made equal to YL:

Vo '5 - = -1/2, (Eq. 5)

Vin and the circuit's output impedance are matched to the line impedance. 10 Accordingly, the delayed transmitted signal Vin appears damped by a factor of 1/2 in the summing circuit (270) of FIG. 3B. If the digital F-filter (272) causes attenuation by a factor of 1/2 and an absolute delay which corresponds to the round trip delay from the input of the H-filter (116) to the output of the post-filter (125), then the output signal of the F filters (272) compensate for the undesired returning transmission signal Vin in the summing circuit (270), so that a voltage Vo appears at its output 15, which does not contain any components of the transmitted signal Vin. The implementation of the F filter (272) is simplified under these conditions, since it can consist of a simple delay element of the type of a shift register. The damping with a factor of 1/2 is caused by a right shift of the binary value of the voltage Vin by one position and rounding off

In view of a more general application of this technique to cases where the immittances of the type of prefilter and the battery supply are not negligible, but are known, gH can be changed so that the stresses on the line by the immittances are eliminated, i. h .: 25 gH = YL-Yx, (Eq. 6) where Υχ corresponds to their combined apparent conductance. Under these circumstances, the round trip gain (via encoder path (108) and decoder path (110)) changes to: 30

Y x (Eq. 7) - = -1/2 + -,

Vin 2Yl 35 and the net apparent conductance at connections (102) and (104) seen from the line remains the same Y ^ The second term in equation 7 means that the H-filter (116) must be modified to compensate for,, and it follows that the F filter (272) is modified in such a way that: 40

(Eq. 8) 45 so that the transmission signal components that come back undesirably to the receiver output at VQ are extinguished. 50 -7- 55

AT 394 295 B

In the representation after the Z transformation, the following applies to the H filter: Z2 + KjZ + K2 N (Z) Η (Z) = Ko- = Ko- Z2 + ZK3 + K4 D (Z) 1 + KjZ'1 + K2Z '2 N (Z_1) (Eq. 9)

Ko --- Ko-, 1 + K3Z'1 + K4Z " 2 DCZ’1) where Z is a delay operator and the values K are programmable coefficients and where N and D are numerator and denominator polynomials. The poles and the zeros of the filter are limited so that they lie within the unit circle of the Z plane.

If the above relates to the formation of XL, this means that there can be any combination of resistors, capacitors and coils that contains no more than two elements that are not resistors. However, this limitation is not a mandatory consequence of the technique, but was chosen only to simplify the shape of the H-filter (116) with regard to its implementation in highly integrated technology, in such a way that a simple recursive digital filter with five coefficients is obtained. In addition, the aforementioned limitation on the number of elements that make up is a good practical approximation of the

Function. For other applications than remote telephone subscriber circuitry, the definition of the H filter (116) can be expanded according to the complexity of the immittances to be taken into account on the basis of the technique described here. For example, gH, which corresponds to Y ^ - Υχ in the event that Υχ contains a component that is not a resistor, can be designed such that this function has third-order polynomials in its numerator and denominator. Thus, it is possible for YL to meet the above limitation to no more than two elements that are not resistors.

The filter coefficient adjustment unit (130) will now be described with reference to FIG. 5. The task of adjusting the coefficients of the H filter (116) so that gH becomes YL (or YL - Y) fulfills this adjustment

Unit (130). The readjustment unit (130) routinely checks under the control of a monitoring system whether the coefficients of the digital filter are set correctly and takes care of their initial setting. The adjustment unit works in the so-called " OFF-Line mode " (not in the time interval of a connection establishment or a connection release). The coefficient adjustment unit can be line-specific or it can be switched on N lines in time-division multiplex operation. Once the filter coefficients are set, they only need to be checked periodically at periodic intervals under the control of the monitoring system, since any given telephone line characteristic does not normally change from one day to the next. This allows the coefficient adjuster to be used by a plurality of subscriber line circuits in time division multiplexing, so that their cost is divided over a number of lines. Fig. 5 shows the conditions under which the adjusting unit according to the invention works. The switches (106) and (107) of Fig. 2 separate the F filter (272) and the feedback via line (112) from the encoder path (108) when they are open. The H filter (116) is in the bridged state (it is effectively shorted between its input and output terminals, i.e. it is gH = 1).

2 can be described as follows:

The encoder path (108) includes an encoder pre-filter (133), a sigma-delta encoder (135) and a filter (137) with decimator and low pass functions. The output signal of the filter (137) and the output signal of the F filter (272) are added to one another in a digital summing circuit (139), and an intermediate receiver output signal is produced on the line (141), which is filtered in a reception filter (143) becomes.

The filter (143) supplies on its output line (145) the four-wire reception signal of the subscriber line circuit. The four-wire transmission signal on line (147) is filtered in a transmission filter (149). The output signal of the filter (149), an intermediate transmit input signal, is fed to the digital filter (116) via a summing circuit (151), where this intermediate transmit signal is added to the signal which is then on the line (112) as Feedback signal appears when the switch (106) is closed under the control of the multiplexer (130). Before decoding in the decoder (114), the filtered signal is filtered again in an interpolating filter (153). For completeness, programmable signal generators (155) are shown, for details of which see U.S. Patent 4,161,633 -8-

AT 394 295 B. The central part (130), which essentially contains the multiplexer (157) and the coefficient adjustment unit (159), has means for distributing control coefficients for 1 to N subscriber line circuits, as generally shown at (157). The central part also contains a digital reference generator (161).

A remote control device at the subscriber end of the line is brought into the state of the loop closure by an operation monitoring action. The reference generator (161) supplies a test signal input with a reference signal with uniform energy distribution within the normal operating frequency band and with vanishing energy outside this band. The output signal 0 (Z) is correct if (Fig. 5) (Eq. 10) E (Z) = O, where Z is the delay operator, and E (Z) = 0 (Z) = R (Z) Z 'L (Eq. 11)

The term Z '^ compensates for the known absolute sampling delays that the signal experiences on the way through the loop. Under these conditions the following applies: N (Z_1) -Ko = YL (Z). (Eq. 12) D (Z_1)

This is the desired result and the resulting coefficients Kq, Kj, K3, K4 have been set correctly and can be loaded into the H filter (116) for normal operation. At the beginning of the readjustment process, the H filter (116) is loaded with test coefficients or with the previously valid coefficients so that the setting can converge quickly. The adjuster solves a set of simultaneous partial differential equations to minimize the mean square G between R (nT - LT) and 0 (nT) as a function of the coefficients K ^. This is represented as: 5G δΟ (ηΤ) - = Σ [(0 (ηΤ) - R (nT - IT)] -, (Eq. 13) 8Kk " 5Kk

According to the well-known theory, for example by Lucky & Rudin in BSTJ, Nov. 1967 and by Weiner in MIT Press, published in 1964, Appendix B, " Time Series Analysis " is described.

The output signals P0 (nT), Pj (nT), etc., represent the partial derivatives 0 (nT) according to the filter coefficients. These output signals are multiplied by E (nT), summed, and the sum is rounded off and yielded at periodic time intervals the output signals Cj ,, which update the

Polynomials N (Z " *) and D (Z_1) with new coefficients allow according to the equation:

Kj ^ new = Kj, old - Cj, A (Eq. 14) where Δ is an increment value or factor for setting the step size. This rounding technique and the means for applying Equation 14 are known in the field of automatic equalizers for non-recursive equalizer structures. 5, however, implements the function of an automatic equalizer with a recursive equalizer structure. 5 supplies the values P ^, which are the -9-

AT 394 295 B

Consider the interaction of the coefficients Kk, which has so far been considered as one of the limiting factors in recursive equalizer structures. This may be one of the reasons that non-recursive equalizer structures have been predominant, because of their inherent simplicity in the partial functions of the automatic equalizers based on the criteria of the mean square of error. Such non-recursive structures of the type known hitherto require 30 to 60 complex coefficients, whereas a recursive structure according to the present invention requires only five coefficients and thus leads to a considerable simplification of the circuit complexity.

5, which shows the mode of operation of the recursive equalizer, it emerged that the relevant equations are the following: N (Z_1) PQ (Z) = A (Z) 1- DiZ'1) (Eq. 15) K0z · 'P! (Z) = A (Z) - DfZ'1) (Eq. 16) P2 (Z) = Z-1P1 (Z) (Eq. 17) -V1 P3 (Z) = -P0® D (Z_1) (Eq . 18) P4 (Z) = Z-1P3 (Z) (Eq. 19) 0 (Z) = K0P0 (Z) (Eq. 20) E (Z) = 0 (Z) - R (Z) (Eq. 21) N Ckneu = Σ Pk (nT). E (nT), rounded. 1 (Eq. 22) K ^ new = Kcold - C ^ new A (Eq. 23)

Using the above equations in an iterative process, the coefficients of the H filter (116) are continuously updated to a point where the values Ck are negligible

Values are:

Ck <5k (Eq. 24) -10-

AT 394 295 B

The values for öjj depend on noise and other factors and are determined empirically. If the criterion according to equation 24 is fulfilled, the filter coefficient adjustment unit has fulfilled its task and can be assigned to another telephone line.

The values obtained for the coefficients K ^ are loaded into the H filter. In the event that Υχ = 0, the transmission characteristic of the F filter is simply 1/2 with an absolute delay which corresponds to the round trip delay caused by the digital filters, which is indicated by Z &quot; L in the representation after the Z transformation can graze. If Fall Φ 0, the equalization process results in: N (Z4) -K0 = YL (Z) + Υχ (Ζ) (Eq. 25) DCZ-1)

To get the required value for gH, the known quantity Yx, multiplied by two, must be subtracted: N (Z_1) gH (Z) = -KQ - 2 Υχ (Ζ) = YL (Z) - YX (Z) ( Eq. 26) D (Zb

In this case, the number of partial radio functions must be increased in order to be able to use third degree polynomials for N (Z'l) and D (Z * ^), which is necessary if a first degree polynomial is to apply to für (Ζ) .

Under these conditions the following applies to the F filter: F =

gH YL-Yx

2YL

2YT (Eq. 27)

This value for the F-filter (272) can be calculated in the operational monitoring system from the equalization results obtained when gH was determined and from the known value for Υχ.

Alternatively, the value of F can also be determined directly. This operation is carried out as follows: The coefficients of the H filter (116) are loaded, the switch (106) in the feedback path (112) of FIG. 2 is closed by a switch control signal from the multiplexer (157), the switch (107) in The path of the F filter (272) is opened and the adjustment process in the adjustment unit (130) starts. This sequence of operations yields: N (Z_1) 2Yl -KG = - (Eq. 28)

D (Z_1) YL-YX and thus is: D (Z_1) 1 F = - (Eq. 29) N (Z ']) K0 - 11 -

AT 394 295 B

The above provides coefficients for a recursive filter of the same type as the H filter (116) with third degree polynomials for N (Z_1) and D (Z_1) with a first degree polynomial for Υχ. The respective circuit implementation of the H and F filters can be done using known technology.

6 shows generalized H and F filters (116) and (272) for processing polynomials with degree k / 2 for N (Z **) and D (Z '^) according to equation 30. γ ( Ζ) κ0 (ΐ + κ1ζ1 + κ2ζ-2 ... Kk / 2z · ^ 2) - = - (Eq. 30) X (Z) (1 + Kk .Z'1 ... KjjZ '^ 2 ) - + 1 2

The coefficients and data are stored in a semiconductor random access memory which is made up of stack memories (300) and (302) and rotating stack memories (304) and (306) in order to facilitate the fetching and storing of the information . At each sampling time T, output data from stack memories (300) through (306) is fed to a multiplier / summing circuit (308) which multiplies the required output signal Yn appearing on line (310) and successively

Accumulate the results calculated according to Equation 31

Yn = Ko (Xn + KiXn.i + K2Xn-2 ... KkXn.k) - 2 2 (Kk Yn-1 + Kk Yn-2 --- KkYn-k) (GL 31) - +1 - +2 - 2 2 2

When switches Sj (312) and S2 (314) are in position 1, the first expression KQXn is calculated. Sj is set in position 3 and the x terms are calculated. After this calculation, S] and S2 are set in position 2 and the Y terms are calculated. Thus, k + 1 multiplication / addition operations are carried out

This can easily be done within the sampling period if the same memory and multiplier (308) and accumulator (316) are used for the H and F filters. Thus, an H and F filter, each with six coefficients (k = 6), are 14 multiplication / addition operations, with approximately one microsecond being permitted for each such operation, so that a sampling period of T of 14 microseconds can be sufficient. Higher-degree polynomials can be processed if the computing and storage operations work a lot more in parallel. Other forms of recursive filters are possible, and the filter structure of Figure 6 is intended to be only one example.

According to the present invention, various designs of the coefficient adjustment unit can be implemented. A preferred exemplary embodiment uses semiconductor read / write memories, an arithmetic and control unit and control logic for digital signal processing using the principles underlying the equations given above. In addition, algorithms other than the algorithm of the mean square error can be used to determine the coefficients of the H and F filters, even if this algorithm serves as the basis for the functioning of the adjusting unit in the exemplary embodiment dealt with here. For example, the algorithm described is based on determining the values of Ck over a period of sampling intervals that is NT. Each time the values for Ck are calculated, the coefficients Kk are updated every N sampling intervals, so that:

mNT

Ck (mNT) = Σ Pk (nT) E (nT), rounded (Eq. 32) n = (m-l) NT + l -12-

AT 394 295 B and the new coefficient values are calculated based on the components Ck of the gradient vector according to equation 33:

Kk (m) = Kk (m-l) -Ck (m) (Eq. 33)

To simplify the calculation of Ck by approximation:

Ck (nT) = Signum Pk (nT). Signum E (nT), (Eq. 34) so the coefficients Kk can be updated after every sampling interval T, which results in a faster

Convergence against the end values is possible and as a result the circuit complexity required can be reduced. This algorithm is a simplification which approximates the performance of the algorithm of a recursive filter based on the mean square error, and which has advantages especially when emphasis is placed on rapid convergence and reduced circuit complexity.

The digital filters used to form the partial output signals Pk can be implemented in the same way for the F and H filters.

A block diagram of a special embodiment of the adjusting unit (159) will now be explained with reference to FIG. 9. An arithmetic and control unit (500) operated in time-division multiplexing carries out successive arithmetic and logic operations with the information stored in the memory (502), under the control of output signals of the control logic (504), which are synchronized with the clock of a clock generator (506) . The clock frequency of the clock generator (506) is in turn synchronized with a data sampling clock signal and is a multiple thereof. The first and second read / write memories (508) and (510), which together form the memory (502), are controlled by a series of control words, each of which also controls the operation of the arithmetic and control unit (500).

External control words on line (512) are shifted from a shift register (513) to control gates (514) to provide control words on line (516) under the control of drive logic (504). Constants can be loaded into the memory (502) under external control, the contents of the memory can be checked externally, and the readjustment process can be started externally. When the readjustment process is closed, a logic output signal EC is emitted.

FIG. 10 shows the structure of the first memory (508), which consists of four stack memories (520, 522, 524 and 526), the functional details of the stack memory (520) being shown.

The stack memories (522, 524 and 526) operate in the same way as the stack memory (520). Each stack (520) through (526) operates such that each new input word on the memory bus lines (528) and (530) replaces the last of the words previously entered into the stack.

This is a so-called &quot; first in - last out &quot; (FILO) mode of operation. Every data place in every stack memory can be directly addressed or read out. The stack (520) stores the words A (n), A (n-1) ..., the stack (522) stores Po (n), Po (nl) ..., the stack (524) stores Pl ( n), Pl (n-1) ... and the stack (526) stores P3 (n), P3 (n-1) .... The decoder (532) decodes the control words from line (516) come, and derives from it the individual control signals for each stack. The output signal of each read process is stored in the memory register (534) and forms an input signal of the arithmetic and control unit (500).

FIG. 11 shows the organization of the second memory (510) (FIG. 9), which contains memory sections (540, 542, 544 and 546). The storage section (540) stores constants Kk, including KQ to K4. The

Memory section (542) stores data words 5k including 6Q to 84. The memory section (544) stores data words R (n) from the reference generator (181) which are stored via a shift register (548). The storage section (546) stores E (n). All store-in processes or all read-out processes in or from all memory sections (n) can be controlled separately from one another by suitable control words which are decoded in the decoder (550). Each addressed data word in each of the memory sections (540 to 546) is entered into a memory register (552) via a gate circuit (554) when it is addressed. The memory section (544) is constructed as a stack, in which each stored data word R (n), R (n-1) ... R (n-1) can be addressed directly. The arithmetic and control unit (500) has as input signals the output signals of the registers (534) of the memory (508) and (552) of the memory (510) (FIG. 9). The arithmetic and control unit performs the arithmetic operations with its input signals and then enters the results into its accumulator (556). The results are then stored and transferred to the memory under the control of the control logic (504). -13-

AT 394 295 B

The actual arithmetic and logical operations performed by the arithmetic and control unit (500) are summarized below:

Multiply function Multiply / add Subtract / add Increment Decrement Complement ACC if negative Zero ACC start

surgery

C (R1). C (R2) - &gt; ACC

C (R1). C (R2) + C (ACC) - &gt; ACC

C (= R1) = C (= R2) + C (ACC) - &gt; ACC C (M2) + LSB- &gt; M2 C (M2) - LSB- &gt; M2 if Signum ACC negative:

2N-C (ACC) - &gt; ACC

Value &quot; 0 &quot; - &gt; ACC

Start of the adjustment process

In the above, LSB means the least significant bit (LSB = least significant bit).

To carry out the multiplication, the contents of the storage registers (534 and 552) are multiplied and the product is stored in the accumulator (556). To carry out the multiplication / addition, the contents of the memory registers (534 and 552) are multiplied and the result is added to the contents of the accumulator (556). To carry out the subtraction / addition, the contents of one or both registers (534 and 552) are added to the contents of the accumulator (556), suitable sign changes being carried out under the control of a field code.

To carry out the incrementation or decrementation, the contents of the respective storage space are increased if the sign of an accumulator content (560) is negative and reduced if this sign is positive, i. H. according to the field code. To perform the function "Complement ACC if neg." the sign of the accumulator content is changed to a positive if it is negative.

To perform the &quot; Implement ACC &quot; the numerical value zero is stored in the accumulator. In order to perform the staging which allows the external loading of constants when an external write control signal is present, the accumulator contents are cleared and the identifier bits of the accumulator at (562) are reset.

The structure of the control word output by the control logic (504), which is input via the line (516) into the memories (534 and 552), has, for example, a 6-bit field for the memory M1, a 7-bit field for the memory M2 and a 5-bit field for the arithmetic and control unit (500). Each control word thus has 18 bits. This breakdown of the tax word can be represented as follows:

Field for Ml 1 r Memory Ml: Enter 2 Address 1 Select II Subtotal 2 6 bit Field for M2 1 r Memory M2: Read / Write 1 Address 3 Select II Subtotal 2 7 Bit -14- AT 394 295 B (continued) Field for Arithmetic unit (500): Operation II subtotal 5 5 bits

Entire control word: field f. Ml field f. M2 field f. Arithmetic || Total 6 7 5 18 bits

FIG. 12 shows a flow diagram of the mode of operation of the adjusting unit. The sequence is as follows:

Step 1

Preparing to start

An external start signal on line (570) resets a flag signal appearing on line (572), which signifies the completion of an adjustment process, and memory locations and the accumulator (556) and the associated registers are deleted when an external write signal is present the line (574) is present, the control logic (504) allows the initial values Kk and 5k to be loaded externally via a shift register (576). If there is no external write signal, the stored values Kk and are provided via the control logic (504).

step 2

Calculation of pofn)

The values A (n) and R (n) are provided from the outside at the sampling times nT. R (n) and A (n) are entered into their memories (520 and 548), A (n-2) is fetched into the register (534) of the memory Ml, and K2 is entered into the register (552) of the memory M2 fetched. Your product is calculated by the arithmetic and control unit (500) and kept in the accumulator (556). Similarly, A (n-1). Kj calculated thereon and added to the content of the accumulator (556). A (n) is then added to the contents of the accumulator. The Po terms to be coupled are subtracted from the accumulator content according to the equation: new Po (n) = A (n) + KjA (nl) + K2A (n-2) - K3Po (n) - K4Po (nl) (Eq. 35) N (Z_1)

This operation corresponds to the function shown in FIG. 5. The result Po (n) is in the for D (Z ~ 1)

Po provided stack memory (522) stored.

Step 3 -1

-KoZ

The term P3 (n) at the output of the filter, which similarly calculates the transfer function according to: (FIG. 5), is D (Z_1) new P3 (n) = -KoPo (nl) - P3 (n) K3 - P3 (n-1) K4 (Eq. 36) -15-

AT 394 295 B

And the result is stored in the stack (526) provided for P3.

Step 4

Calculation of PjCn)

Similarly, Pj (n) is calculated and placed in the stack P | stored according to: new Pj (n) = KoA (n-l) - Pj (n). K3 - Pj (n-1). K4, (Eq. 37) KöZ'1 which corresponds to the filter function shown in Fig. 5. D (Z-1)

Step 5

Calculation of E (rri

The fault antenna is calculated according to: (Eq. 38) E (n) = Po (n) Ko + R (n-1) E (n) is retained, i.e. H. saved in the specified place. This operation corresponds to the function of the summing circuit shown at the top right in FIG. 5.

Steps 6 and 7

Calculate the values Cj .. Update the values Kj.

The values Ck are calculated based on the approximation described above. This operation involves incrementing or decrementing Kk based on the sign of the components Ck of the gradient vector, i. that is, if the sign is negative: increase Kk

Ck (n) = Pk (n) E (n) (Eq. 39) with a positive sign: Kj, decrease

Steps 8.9 and 10

Check, oh adjustment process completed

The absolute amount of Ck is calculated by changing the sign of Ck if it is negative. The corresponding value is subtracted from the absolute amount of Ck. If the difference in the accumulator is positive, the indicator is set to 1. If the difference is negative, the flip-flop provided for the label remains unchanged, i. i.e., = +: set indicator ICk '- \ = -: leave indicator reset

This step is vectored for each component of the gradient vector Ck. H. for CQ to C4. At the end of this procedure the value of the label is checked and if the label is 0, which means that no component Ck has exceeded the corresponding value 5k, the readjustment process is complete. However, if the indicator is 1, indicating that one or more values of Ck have exceeded their corresponding value 5k, then the readjustment process has not been completed, and

AT 394 295 B the cycle must be repeated. Drive logic (504) returns to step 2 to await the next strobe signal that occurs after an interval T after the previous strobe signal. If the flag is 0, the signal EC indicating the completion of an adjustment operation is output for external detection and the procedure is completed so that the external system can read the values of KQ bis from the memory for external use. At this point, the adjustment circuit can then be assigned to another subscriber line.

As far as the execution times are concerned, the adjusting unit must complete steps 2 to 9 in a time that is less than or equal to the sampling interval T. In the currently customary sampling, the sampling interval is 125 microseconds, which corresponds to a sampling frequency of 4 kHz. Assuming that a maximum of 50 control words are to be processed during steps 2 to 9, each control word must be processed in approximately 2 microseconds. In the worst case, a control word means fetching two words from memory (508) and memory (510), multiplying these words and adding their product to the content of the accumulator. In this case, the following requirements for execution times can be set:

Memory access: 0.5 microseconds

Multiply: 1.0 microseconds

Add: 0.5 microseconds

Total: 2.0 microseconds

These requirements require parallel transfers between the various registers and arithmetic operations in parallel. With a 13-bit arithmetic that meets the telephony requirements, these features can be achieved with today's technology of the highly integrated circuits (LSI) if the equalizer structure described here is used.

Currently commercially available and general purpose microcomputers that can process 8 to 16 bits in parallel have not been able to meet the above performance requirements when using standard programming. However, the new features of the structure of an adjusting unit described here make it possible to meet these performance requirements. A brief summary of these new features results in the following list, among others: (1) memory division into several memory sections that can be addressed simultaneously, (2) special memory organization that facilitates the required operations (directly addressable stack memory), (3) the ability of parallel arithmetic with respect to multiplication and addition, (4) micro-coded control words which simultaneously control the memories and the arithmetic control system, (5) control words which are directly related to the specific operation required, e.g. B. Multiply / add, complement ACC, if neg., Increment, decrement.

An alternative embodiment of the adjustment unit described above can be realized by using a general-purpose signal processor that has special skills in arithmetic operations and memory operations. A simplified block diagram of such a general-purpose processor is shown in FIG. 7.

The circuits described can all be implemented using digitally highly integrated technology. Adding the feedback and feedforward loops and the F and H filters and replacing the standard decoder output amplifier with a slope amplifier makes it possible to implement a complete subscriber line circuit on a single, highly integrated device. The F and H filters are simple recursive filters that can be installed in a codec module or in a codec filter module. Thus, the present invention eliminates the need for a prior art two-wire, four-wire analog circuit and discrete termination and matching networks because they can be replaced with the programmable digital, highly integrated circuits described above. This leads to lower costs in production, installation and operational monitoring and also brings about an improvement in the performance characteristics.

If non-recursive filters can be used as H and F filters, their costs would be higher than with the recursive filters described.

A prior art non-recursive filter coefficient adjuster could also have been used for either non-recursive or recursive F and H filters, but this would be less favorable than the solution described above. The non-recursive filter structures, which are provided by a non-recursive adjustment unit, could be converted into a recursive structure of the type described if the algorithm of Fletcher-Powell is used. This is described by Deczky: &quot; Synthesis of -17-

AT 394 295 B

Recursive Digital Filters &quot;, IEE Trans. Audio Electro Acoust. Vol.AU-20, Oct. 1972, pages 257 to 263. But this technique also requires a considerably greater circuit complexity than the completely recursive structure according to the present invention. The memory (330) according to FIG. 7 contains various memory sections, the memory section (336) for storing the coefficients C ^, the memory section (338) for storing coefficients Kj., The memory section (340) for storing correlation sums the memory section (342 ) for storing the values and δ ^, the storage section (344) for storing intermediate results A (n), Ρ ^ (η), ... Ρ ^ η__y and the storage section (346) for storing a control program to access the stored data according to the addresses provided by a control logic (332). The calculation is carried out using an arithmetic unit (334).

Figure 8 shows a generalized digital two-wire to four-wire converter for full duplex signals on line (400).

The encoder has a gain of 1 and contains a pre-filter (402) for the analog-digital converter (403) and a digital post-filter (406). The circuit of Fig. 8 operates in the same manner as that of Fig. 3B, and the impedance matching to the telephone line is similar. The decoder path, which contains a digital-to-analog converter (406), an associated pre-filter (408) and an associated post-filter (410), converts the transmission signals coming from the line (412) into analog signals which are transmitted via the line (400) be sent out. A slope amplifier (414) ensures an infinite output admittance. The automatic coefficient adjuster (130) (Fig. 2) provides updated filter coefficients and control signals for the H filter (416) and the F filter (418) so that undesired reflux signals are suppressed at the output of a summing circuit (420). The feedback encoder and F filter (418) are turned on by switches (422 and 424), these two switches being controlled by the adjuster (130) (Fig. 2).

Multiplexer (157) (which is generally described in FIG. 2) is shown in more detail in FIG. 13. The multiplexer makes it possible to use the adjustment unit in time division multiplex for a large number of lines 1 to N. The multiplexer (157) determines which line is to be connected to the adjustment unit. It processes the signals A (n) from the plurality of subscriber line circuits in time division multiplex and distributes the filter coefficients, the switch control signals and the control signals for the operating modes of the filters and the reference signal of the reference generator (161) to the line which is selected in each case under the control of the operation monitoring system.

Signals R (n) from the digital reference generator (161) reach lines 1 to N via a gate circuit (600), with R (n) being controlled by AND circuits ADD 1 to ADD N, which are controlled by the operation monitoring system via logic (602) The appropriate address is shifted through a shift register (604), decoded in a decoder (606) and, as shown, supplied to the logic (600).

The clock signals and the control signals and the operating mode of the filter, which come from the operational monitoring system, are fed to lines 1 to N via a gate circuit (608). These signals are stored in flip-flops (610 and 612), the output signals of which are combined with the signals ADD 1 to ADD N by the decoder (606) in AND circuits. The words A (n) from lines 1 to N are selected by a gate circuit (614) and fed to the adjuster circuit as the output of an OR circuit (616). Each input signal A (n) from each of the lines 1 to N is combined with signals ADD 1 to ADD N from the decoder (606) in AND circuits of the gate circuit (614).

The filter coefficients of the F and H filters for each of the lines 1 to N, for example for the F filter (272) and the H filter (116) (FIG. 2) of the subscriber line circuit 1 from N, are determined by the adjusting unit via the gate circuit (618) fed to line 1 from N. For this purpose, the filter coefficients run through the logic circuit (618) AND circuits, at the other inputs of which the signals ADD 1 to ADD N from the decoder (606) are located and thus reach the correct line.

Finally, Fig. 14 is considered, which shows a block diagram of the most important interfaces connections between the adjustment unit, the operational monitoring system, the multiplexer and the subscriber line circuits. The operational monitoring system (650) can consist of a conventional data source, for example a computer and the associated memory. The data signals and control signals shown, which are described elsewhere, ensure the necessary timing and the cooperation between the adjusting unit, the operational monitoring system and the subscriber line circuits on the way via the multiplexer (157). -18-

Claims (12)

AT 394 295 B PATENT CLAIMS 1. Interface device for connecting analog, full-duplex telephone lines to a digital exchange, which consists of a group of two-wire to four-wire converters, each of which is connected to a single remote line and an analog-digital converter in the Contains four-wire reception path and a digital-to-analog converter in the four-wire transmission path as well as digital filters, the converters effecting a resistance adaptation to the subscriber lines, characterized in that in the two-wire four-wire converters (100) the one input of a digital summing circuit ( 151) is connected to the output of the analog-to-digital converter (135) in the four-wire reception path, which converts the analog signal received by the telephone line (102, 104) into a composite digital signal, and the other input to the transmission input of the digital exchange is connected, the digital output signal of the analog-digital Wan dlers (135) is added as a negative feedback signal to the digital transmission signal coming from the digital exchange that the output of the digital summing circuit (151) with the input of a recursive digital filter (116) with adjustable filter coefficients for the direct setting of a predetermined output impedance curve, is connected that the output of this digital filter (116) is connected to the input of the digital-to-analog converter (114), whose analog output signal voltage, which represents the information signal to be transmitted via the telephone line, a device (200) for converting the analog signal voltage into an analog current, which is fed to the telephone line (102, 104) and to the input of the analog-digital converter (135), this device having a high output impedance compared to the adapted telephone line and that of the digital filter (116 ) has the set impedance, that the interface device further includes a central part (130) common to the group of two-wire to four-wire converters, and that the central part (130) contains a multiplexer (157) and a filter coefficient adjustment unit (159), the multiplexer comprising the filter coefficient adjustment unit connects to each of the two-wire four-wire converters of the group in time-division multiplexing mode and adjusts the filter coefficient adjustment unit by minimizing an error between the digital received signal that the two-wire four-wire converter outputs to the digital exchange and a reference signal that adjusts the filter coefficients for automatic impedance matching. 2. Device according to claim 1, characterized in that the device (200) for converting the analog signal voltage into an analog current consists of a steepness amplifier (Fig. 4). 3. Device according to claim 1 or 2, characterized in that the filter coefficients of the digital filter (116) are adjusted by the filter coefficient adjustment unit (159) in an adaptive manner. 4. Device according spoke 1, 2 or 3, characterized in that the recursive digital filter (116) has five adjustable filter coefficients (KQ to K4, Fig. 5). 5. Device according to one of claims 1 to 4, characterized in that in each two-wire four-wire converter between the coding path and decoding path, a second digital filter (272) with adjustable coefficients is connected, to which the digital transmission signal coming from the digital exchange is supplied and which, by multiplying its input signal by the coefficients, outputs a signal representing the undesired portion of the received signal originating from the transmitted signal, the adjustable coefficients of this filter being derived from the filter coefficients set in the filter coefficient adjusting unit (159), and a digital summing circuit ( 139), the output signal of the second digital filter (272) is combined with the composite digital output signal of the analog-digital converter (135) in such a way that its unwanted echo components originating from the transmission signal are compensated for. 6. Device according to claim 5, characterized in that the second digital filter (272) is a recursive digital filter. 7. Device according to one of claims 1 to 6, characterized in that the analog-digital converter (135) is a sigma-delta encoder. -19- AT 394 295 B 8. Device according to claim 7, characterized in that the analog-to-digital converter (135) contains a pre-filter (133) for limiting its input bandwidth and a post-filter (137) for setting a desired feedback amplifier curve. 9. Device according to one of claims 1 to 8, characterized in that a switch (106) controlled by the central part (130) is provided in the feedback path between the output of the analog-digital converter and the digital summing circuit connected upstream of the first digital filter (116) is. 10. Device according to one of claims 1 to 9, characterized in that the central part in the form 10 of a single highly integrated circuit and each two-wire-four-wire converter (100) is realized in the form of a single highly integrated circuit. 11. Device according to one of claims 1 to 10, characterized in that the filter coefficient adjustment unit (159) by digital processing of the input signal and the reference signal calculates a digital 15 gradient vector with the components CQ to Cj, and the coefficients of the recursive filter depending on adaptively adjusts the calculated gradient vector in a rapidly converging process such that the digital filter operates recursively and emits an equalized output signal after convergence. 20 To that 12 sheets of drawings' i J -20-